Graphics data is typically stored in memory in a one-dimensional (1D), or linear, fashion, although data is often conceptually depicted in two-dimensions (2D). For example, a texture map is conceptually represented as having two-dimensions (u, v), but in actuality, the texture data is stored in memory such that the memory addresses for the texture map data are sequential. In accessing the texture data, the requested memory addresses are incremented to access the texture data in a row-like fashion. When the address for the texture data for the last texel in a row is accessed, the memory address is again incremented to roll over and access the texture data for the first texel in the next row.
Typically, data that is requested for processing is retrieved from memory in blocks and temporarily stored for processing in memory pages. In the case of a texture map, each memory page typically includes the texture data for one row of the texture map. When additional data is requested for processing, the memory page of data is first checked for the requested data before performing a memory access operation. If the requested data is already present in the memory page, then overall processing time is reduced. However, if the requested data is not present in the memory page, a “page miss” occurs, and another block of data including the requested data is retrieved from memory to replace the former memory page of data. Page misses add to the overall processing time, and consequently, it is desirable to avoid frequent page misses.
A problem with storing 2D data, such as texture data, in a 1D arrangement is that when bilinear interpolation is performed to determine the color value of a destination pixel in a graphics image on which the texture map is applied, the texture data for four texels from two different rows are required. In bilinear filtering, the color values of four texels closest to the respective location of the pixel are weighted and a resulting color value for the pixel is interpolated from the color value of the four texels. The four texels closest to the destination pixel are naturally arranged in a two-by-two square, and are thus occupy two different rows. As previously discussed, the texture data for two different rows are located on two different memory pages. Consequently, when the texture data for the four closest texels to the destination pixel are retrieved, an average of two page misses will occur: once when the data for the two texels from the first row are retrieved, and once again when the data for the two texels from the second row are retrieved. The page misses slow down the processing of the texture data because the data from one page is purged when the texture data of another page is loaded.
An approach to reducing the occurrences of page misses where bilinear interpolation is applied is to implement a “two-dimensional” memory array such that the memory is conceptually segmented into several small 2D arrays. In this fashion, although the width of a texture map is divided into several 2D segments, the texture data for texels of several adjacent rows may be stored on a common memory page. Thus, the number of page misses occurring during texture application is reduced. Another approach to reducing page misses is disclosed in co-pending U.S. patent application Ser. No. 09/515,246, entitled METHOD AND SYSTEM FOR ADDRESSING GRAPHICS DATA FOR EFFICIENT DATA ACCESS to Peterson, filed on Feb. 29, 2000. The approach described therein maps texture data of a texture map into virtual two-dimensional memory arrays but is implemented in a one-dimensional memory space. An offset value is determined from the (u, v) texel coordinates for the requested texels, and from the offset values, a virtual 2D memory address is calculated and used to effectively store in a 1D memory space the texture data in virtual 2D arrays. As a result, where texels of two different rows of the texture map are required for bilinear filtering, the arrangement of the texels in the virtual 2D memory arrays facilitates texel data processing and minimizes the occurrences of page misses. Although the methods described solve much of the problem with page misses, data is mapped according to a 2D arrangement regardless of the type of data being stored. In cases where it is more efficient to store the data in a 1D fashion, the use of the virtual 2D addressing actually creates inefficient data storage.